International Business Machines Corporation

Bo Zhang

Alexander Sobran

Bradley C Herrin

Xianjun Zhu

ChunHui Yang Higgins

Lawrence H Koved

John H Walczyk, Iii

Joseph Paul Hughes

Guilherme Ferreira

Liwei Wang

William Patrick Higgins

Pu Yang

Ke Zhu

Chul Sung

David Wehr

Eleanor Pence

Halley Fede

Chandramouli Visweswariah

Rui Zhang

Ian M Molloy

Tarun Kumar

Mark Steven Squillante

Hagai Aronowitz

Horst Cornelius Samulowitz

Mayank Sharma

Gregory Bramble

Nancy Anne Schipon

Chuan Ran

Parikshit Ram

Didier Boullery

Laura Grace Ellis

Yongchang Cui

Kazutoshi Yamazaki

Blake McGregor

Yingdong Lu

Angela Schoergendorfer

Adam M Gunther

Peter M Van De Ven

Yogesh Rane

Dan Yang

Janhavi Das

Darrough West

Isabella Yamin

Software for producing and verifying computational determinations using a distributed ledger, by: (i) receiving an indication of a first artificial intelligence (AI) inferencing event, the first AI inferencing event including a first AI inferencing result produced by a first machine learning model based, at least in part, on a first input from a user; (ii) computing a hash of the first machine learning model using a cryptographic hash function; (iii) sending a record of the first AI inferencing event to a verification system, the record of the first AI inferencing event including the hash of the first machine learning model; and (iv) receiving a verification from the verification system indicating that the hash of the first machine learning model matches a hash of a second machine learning model and that the record of the first AI inferencing event has been stored in a first distributed ledger.

Producing and verifying computational determinations using a distributed ledger

A system that provides a mathematical formulation for new problem of model validation and model selection in presence of test data feedback. The system comprises a memory that stores computer-executable components. A processor, operably coupled to the memory, executes the computer-executable components stored in the memory. A selection component selects a metric of performance evaluation accuracy; and a configuration component configures performance evaluation schemes for machine learning algorithms. A characterization component employs a supervised learning-based approach to characterize relationship between the configuration of the performance evaluation scheme and fidelity of performance estimates; and an optimization component that optimizes accuracy of the machine learning algorithms as a function of size of training data set relative to size of validation data set through selection of values associated with the configuration parameters.

Methods for automatically configuring performance evaluation schemes for machine learning algorithms

Methods and systems of evaluating a performance of an entity are described. A processor may obtain first data indicating tier attributes of resources of the entity, second data indicating function attributes of the resources, and third data indicating productivity attributes of the resources. The processor may train a model based on the first data, the second data, and the third data, the model may represent transitions of the resources over time. The processor may receive a set of controls including at least an objective to optimize a performance of the entity. The processor may generate a controlled model by integrating the set of controls into the model. The processor may determine a set of outcomes from the controlled model that includes at least a set of transitions relating to the resources that may optimize the performance of the entity.

Performance evaluation based on resource dynamics

In an approach to quarantining source code to prevent confidential information exposure, one or more computer processors store a pushed codebase associated with a user to an isolated quarantine area, wherein access to the quarantine area is restricted to the user through user interface code visibility enforcement and protocol code visibility enforcement. The one or more computer processors dynamically adjust a timeout period based on codebase complexity, user preferences, associated dependencies, codebase size, minimum execution temporal period for one or more scans, and system specifications. The one or more computer processors responsive to identifying one or more instances of confidential information, perform one or more remedial actions.

Dynamically adjusted timeout quarantined code scanning

Provided are a computer program product, system, and method for classifying code as introducing a bug or not introducing a bug to train a bug detection algorithm. For each commit in a commit history of code changes to the code base, a determination is made of lines of code changed by the commit. For each line of code changed by the commit, a determination is made as to whether the commit is for a bug fix. A determination is made as to whether a previous commit changing the line of code changed by the commit for the bug fix in response to determining that the commit is for the bug fix. Indication is made that the previous commit introduced a bug. The algorithm is trained to classify changes to lines of code by commits indicated as having introduced a bug as bug introducing commits.

Classifying code as introducing a bug or not introducing a bug to train a bug detection algorithm

In an approach, one or more computer processors create a dictionary for each source code commit in a set of historical source code commits associated with a software deployment; create a similarity model based on the created dictionary for each source code commit in the set of historical source code commits; generate a vector embedding for a source code commit pair based on a set of log differences between source code commit pairs utilizing the created similarity model; generate, responsive to a new source code commit, a new vector embedding based on a set of log differences between the new source code commit and a preceding source code commit utilizing the created similarity model; generate a defect likelihood utilizing the generated new vector embedding; determine, responsive to the generated defect likelihood exceeding a defect likelihood threshold, that the new source code commit contains defects.

Vector embedding of relational code sets

A computer-implemented method, system and computer program product for representing source code in vector space. The source code is parsed into an abstract syntax tree, which is then traversed to produce a sequence of tokens. Token embeddings may then be constructed for a subset of the sequence of tokens, which are inputted into an encoder artificial neural network ('encoder') for encoding the token embeddings. A decoder artificial neural network ('decoder') is initialized with a final internal cell state of the encoder. The decoder is run the same number of steps as the encoding performed by the encoder. After running the decoder and completing the training of the decoder to learn the inputted token embeddings, the final internal cell state of the encoder is used as the code representation vector which may be used to detect errors in the source code.

Representing source code in vector space to detect errors

Customizing computing resource allocation based on machine learning is provided. A plurality of current users of a service are evaluated based on attributes of each current user. A user value score is generated for each current user based on the attributes. The plurality of current users is clustered into a plurality of user groups based on the user value score. A likelihood of each prospective user becoming a paying customer is predicted based on collected behavior data corresponding to each prospective user. A likelihood of an application corresponding to the service and associated with a particular user will misbehave is predicted based on resource usage and features of the application. Resource allocation to the application corresponding to the service and associated with the particular user is customized based the likelihood that the application will misbehave and whether the particular user is a current user or a prospective user.

Machine learning-based resource customization to increase user satisfaction

Software for producing and verifying computational determinations using a distributed ledger, by: (i) receiving a first input from a user; (ii) producing a first computational determination utilizing a first computational model, wherein the first computational determination includes a first computational output that is based, at least in part, on the first input; (iii) computing a hash of the first computational model; (iv) sending a record of the first computational determination to a verification system, wherein the record of the first computational determination includes the hash of the first computational model; (v) receiving a verification from the verification system indicating that the hash of the first computational model matches a hash of a second computational model and that the record of the first computational determination has been stored in a first distributed ledger; and (vi) in response to receiving the verification, providing the first computational output to the user.

Approaches presented herein enable domain knowledge-based software defect prediction. More specifically, a cost function configured to train a machine learning model to predict a defect in a software version is obtained. A domain knowledge penalty metric is then determined for a software development team associated with the software version. A linear combination of the determined domain knowledge penalty metric is then applied to the obtained cost function to create a domain knowledge-modified cost function. Using this domain knowledge-modified cost function, the machine learning model is then trained based on a minimization of the domain knowledge-modified cost function. Once trained, the machine learning model is used to report a set of predicted values based on the domain knowledge-modified cost function trained machine learning model.

Devops driven cognitive cost function for software defect prediction

A method, system and computer program product for obtaining vector representations of code snippets capturing semantic similarity. A first and second training set of code snippets are collected, where the first training set of code snippets implements the same function representing semantic similarity and the second training set of code snippets implements a different function representing semantic dissimilarity. A vector representation of a first and second code snippet from either the first or second training set of code snippets is generated using a machine learning model. A loss value is generated utilizing a loss function that is proportional or inverse to the distance between the first and second vectors in response to receiving the first and second code snippets from the first or second training set of code snippets, respectively. The machine learning model is trained to capture the semantic similarity in the code snippets by minimizing the loss value.

Generating vector representations of code capturing semantic similarity

An example operation may include one or more of seeking consensus among users of a source tree to approve execution of a blocked command to a source control system, creating a child ledger for a user issuing the blocked command and initiating a blockchain transaction to link the child ledger to a master ledger, merging the child ledger into the master ledger when consensus is reached, and executing the blocked command.

Blockchain technique for immutable source control

Training data identifying a plurality of pairs is received. Each pair identifies one or more separate code snippets known to resolve a respective issue of a plurality of issues. For each pair of the plurality of pairs, a respective issue representation of core content of the respective issue and a linear expression of the one or more separate code snippets in a respective code representation is constructed and a model to correlate the respective code representation as resolving the respective issue representation is trained. The model is queried with a new issue and a selected one of the one or more separate code snippets. The model returns a classification indicating whether the selected one of the one or more separate code snippets is likely to resolve the new issue.

Code recommender for resolving a new issue received by an issue tracking system

An example operation may include one or more of intercepting a command from a user to modify a source tree in a source control system, creating a child ledger associated with a master ledger when the intercepted command is destructive, seeking consensus among users of the source tree to approve execution of the intercepted command, merging the child ledger into the master ledger with a transaction describing the intercepted command, a commit tree history, and status of the consensus, and a tree hash, and merging execution results of the intercepted command into a history of the source tree.

Software for immutably storing computational determinations using distributed ledgers. The software performs the following operations: (i) receiving an indication that a first computational model is ready to be deployed; (ii) storing a copy of the first computational model in a first distributed ledger; (iii) computing a hash of the first computational model using a cryptographic hash function; (iv) receiving an indication that a second computational model has been used to produce a first computational determination; (vi) receiving a hash of the second computational model; and (vii) in response to determining that the hash of the second computational model matches the hash of the first computational model, storing a record of the first computational determination in a second distributed ledger, wherein the record of the first computational determination identifies the second computational model as being the first computational model and includes the hash of the first computational model.

Growing community of inventors

Cary, NC, United States of America

Bo Zhang